Well, it looks like I’m an extreme outlier on this one, because I actually voted it down because I thought it got a lot wrong, and for bad reasons.
First of all, despite criticizing EY for “needing” things that would merely be supercool, matt lists a large number of things that would also be merely supercool: it just doesn’t seem like you need all of those chance values either.
Second, matt seemed to miss why EY was asking for all of that information: that presenting a “neato trick” that happens to work, provides very little information as to why it works, and when it should be used, etc. EY had explained that he personally went through such an experience and described what is lacking when you don’t provide the information he asked for.
In short, EY provided very good reasons why he should be skeptical of just trying every neato trick, matt said very little that was responsive to his points.
despite criticizing EY for “needing” things that would merely be supercool, matt lists a large number of things that would also be merely supercool
Yah, good point—those are meant to be discussion points, but that’s not really very clear as written. I don’t mean to imply that we need everything in the lists, but to characterize the sort of thing we should be looking for.
Second, matt seemed to miss why EY was asking for all of that information
No, I don’t think that’s right. Eliezer is presenting as needful lots of stuff that he’s just not going to get. That seems to be leading him not to try anything until he finds something that passes through his very tight filter. I’m claiming that the relevant filter should be built on expected utility, and that there is pretty good information available (most of the stuff in the lists can at least be estimated with little time invested) that would lead him to try more hacks than the none likely to pass his filter.
EY provided very good reasons why he should be skeptical of just trying every neato trick
I’m very not suggesting that you should try “every neato trick”. I am suggesting that high expected utility is a better filter than robust scientific research. If you have robust research available you should use it. When you don’t, have a look through my lists and see whether it’s worth trying something anyway. You might manage a win.
The upvotes / comment ratio here is remarkably high. What does that mean?
Well, it looks like I’m an extreme outlier on this one, because I actually voted it down because I thought it got a lot wrong, and for bad reasons.
First of all, despite criticizing EY for “needing” things that would merely be supercool, matt lists a large number of things that would also be merely supercool: it just doesn’t seem like you need all of those chance values either.
Second, matt seemed to miss why EY was asking for all of that information: that presenting a “neato trick” that happens to work, provides very little information as to why it works, and when it should be used, etc. EY had explained that he personally went through such an experience and described what is lacking when you don’t provide the information he asked for.
In short, EY provided very good reasons why he should be skeptical of just trying every neato trick, matt said very little that was responsive to his points.
Yah, good point—those are meant to be discussion points, but that’s not really very clear as written. I don’t mean to imply that we need everything in the lists, but to characterize the sort of thing we should be looking for.
No, I don’t think that’s right. Eliezer is presenting as needful lots of stuff that he’s just not going to get. That seems to be leading him not to try anything until he finds something that passes through his very tight filter. I’m claiming that the relevant filter should be built on expected utility, and that there is pretty good information available (most of the stuff in the lists can at least be estimated with little time invested) that would lead him to try more hacks than the none likely to pass his filter.
I’m very not suggesting that you should try “every neato trick”. I am suggesting that high expected utility is a better filter than robust scientific research. If you have robust research available you should use it. When you don’t, have a look through my lists and see whether it’s worth trying something anyway. You might manage a win.
Maybe it means the post was upvoted for agreement, and people don’t have much to add, and don’t want to just say “yay! good post!”?
Could there be a connection to the recent slowing of the rate of new posts to LW?